2008-10-28 00:27:55 +08:00
|
|
|
/*
|
|
|
|
* Copyright (C) 2005, 2006
|
2009-06-14 22:23:09 +08:00
|
|
|
* Avishay Traeger (avishay@gmail.com)
|
2008-10-28 00:27:55 +08:00
|
|
|
* Copyright (C) 2008, 2009
|
|
|
|
* Boaz Harrosh <bharrosh@panasas.com>
|
|
|
|
*
|
|
|
|
* This file is part of exofs.
|
|
|
|
*
|
|
|
|
* exofs is free software; you can redistribute it and/or modify
|
|
|
|
* it under the terms of the GNU General Public License as published by
|
|
|
|
* the Free Software Foundation. Since it is based on ext2, and the only
|
|
|
|
* valid version of GPL for the Linux kernel is version 2, the only valid
|
|
|
|
* version of GPL for exofs is version 2.
|
|
|
|
*
|
|
|
|
* exofs is distributed in the hope that it will be useful,
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
* GNU General Public License for more details.
|
|
|
|
*
|
|
|
|
* You should have received a copy of the GNU General Public License
|
|
|
|
* along with exofs; if not, write to the Free Software
|
|
|
|
* Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
|
|
|
|
*/
|
|
|
|
|
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h
percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files. percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.
percpu.h -> slab.h dependency is about to be removed. Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability. As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.
http://userweb.kernel.org/~tj/misc/slabh-sweep.py
The script does the followings.
* Scan files for gfp and slab usages and update includes such that
only the necessary includes are there. ie. if only gfp is used,
gfp.h, if slab is used, slab.h.
* When the script inserts a new include, it looks at the include
blocks and try to put the new include such that its order conforms
to its surrounding. It's put in the include block which contains
core kernel includes, in the same order that the rest are ordered -
alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
doesn't seem to be any matching order.
* If the script can't find a place to put a new include (mostly
because the file doesn't have fitting include block), it prints out
an error message indicating which .h file needs to be added to the
file.
The conversion was done in the following steps.
1. The initial automatic conversion of all .c files updated slightly
over 4000 files, deleting around 700 includes and adding ~480 gfp.h
and ~3000 slab.h inclusions. The script emitted errors for ~400
files.
2. Each error was manually checked. Some didn't need the inclusion,
some needed manual addition while adding it to implementation .h or
embedding .c file was more appropriate for others. This step added
inclusions to around 150 files.
3. The script was run again and the output was compared to the edits
from #2 to make sure no file was left behind.
4. Several build tests were done and a couple of problems were fixed.
e.g. lib/decompress_*.c used malloc/free() wrappers around slab
APIs requiring slab.h to be added manually.
5. The script was run on all .h files but without automatically
editing them as sprinkling gfp.h and slab.h inclusions around .h
files could easily lead to inclusion dependency hell. Most gfp.h
inclusion directives were ignored as stuff from gfp.h was usually
wildly available and often used in preprocessor macros. Each
slab.h inclusion directive was examined and added manually as
necessary.
6. percpu.h was updated not to include slab.h.
7. Build test were done on the following configurations and failures
were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my
distributed build env didn't work with gcov compiles) and a few
more options had to be turned off depending on archs to make things
build (like ipr on powerpc/64 which failed due to missing writeq).
* x86 and x86_64 UP and SMP allmodconfig and a custom test config.
* powerpc and powerpc64 SMP allmodconfig
* sparc and sparc64 SMP allmodconfig
* ia64 SMP allmodconfig
* s390 SMP allmodconfig
* alpha SMP allmodconfig
* um on x86_64 SMP allmodconfig
8. percpu.h modifications were reverted so that it could be applied as
a separate patch and serve as bisection point.
Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 16:04:11 +08:00
|
|
|
#include <linux/slab.h>
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
#include <asm/div64.h>
|
2008-10-28 00:27:55 +08:00
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
#include <scsi/osd_ore.h>
|
2008-10-28 00:27:55 +08:00
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
#define ORE_ERR(fmt, a...) printk(KERN_ERR "ore: " fmt, ##a)
|
2009-12-16 01:34:17 +08:00
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
#ifdef CONFIG_EXOFS_DEBUG
|
|
|
|
#define ORE_DBGMSG(fmt, a...) \
|
|
|
|
printk(KERN_NOTICE "ore @%s:%d: " fmt, __func__, __LINE__, ##a)
|
|
|
|
#else
|
|
|
|
#define ORE_DBGMSG(fmt, a...) \
|
|
|
|
do { if (0) printk(fmt, ##a); } while (0)
|
|
|
|
#endif
|
|
|
|
|
|
|
|
/* u64 has problems with printk this will cast it to unsigned long long */
|
|
|
|
#define _LLU(x) (unsigned long long)(x)
|
|
|
|
|
|
|
|
#define ORE_DBGMSG2(M...) do {} while (0)
|
|
|
|
/* #define ORE_DBGMSG2 ORE_DBGMSG */
|
|
|
|
|
2011-08-07 10:22:06 +08:00
|
|
|
MODULE_AUTHOR("Boaz Harrosh <bharrosh@panasas.com>");
|
|
|
|
MODULE_DESCRIPTION("Objects Raid Engine ore.ko");
|
|
|
|
MODULE_LICENSE("GPL");
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
static u8 *_ios_cred(struct ore_io_state *ios, unsigned index)
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
{
|
|
|
|
return ios->comps->comps[index & ios->comps->single_comp].cred;
|
|
|
|
}
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
static struct osd_obj_id *_ios_obj(struct ore_io_state *ios, unsigned index)
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
{
|
|
|
|
return &ios->comps->comps[index & ios->comps->single_comp].obj;
|
|
|
|
}
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
static struct osd_dev *_ios_od(struct ore_io_state *ios, unsigned index)
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
{
|
|
|
|
return ios->comps->ods[index];
|
|
|
|
}
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
int ore_get_rw_state(struct ore_layout *layout, struct ore_components *comps,
|
|
|
|
bool is_reading, u64 offset, u64 length,
|
|
|
|
struct ore_io_state **pios)
|
2008-10-28 00:27:55 +08:00
|
|
|
{
|
2011-08-07 10:26:31 +08:00
|
|
|
struct ore_io_state *ios;
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
|
|
|
/*TODO: Maybe use kmem_cach per sbi of size
|
2010-01-28 17:46:16 +08:00
|
|
|
* exofs_io_state_size(layout->s_numdevs)
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
*/
|
2011-08-07 10:26:31 +08:00
|
|
|
ios = kzalloc(ore_io_state_size(comps->numdevs), GFP_KERNEL);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
if (unlikely(!ios)) {
|
2011-08-07 10:26:31 +08:00
|
|
|
ORE_DBGMSG("Failed kzalloc bytes=%d\n",
|
|
|
|
ore_io_state_size(comps->numdevs));
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
*pios = NULL;
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
|
2010-01-28 17:46:16 +08:00
|
|
|
ios->layout = layout;
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
ios->comps = comps;
|
2010-11-17 02:09:58 +08:00
|
|
|
ios->offset = offset;
|
|
|
|
ios->length = length;
|
|
|
|
ios->reading = is_reading;
|
|
|
|
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
*pios = ios;
|
|
|
|
return 0;
|
2008-10-28 00:27:55 +08:00
|
|
|
}
|
2011-08-07 10:22:06 +08:00
|
|
|
EXPORT_SYMBOL(ore_get_rw_state);
|
2008-10-28 00:27:55 +08:00
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
int ore_get_io_state(struct ore_layout *layout, struct ore_components *comps,
|
|
|
|
struct ore_io_state **ios)
|
2010-11-17 02:09:58 +08:00
|
|
|
{
|
2011-08-07 10:26:31 +08:00
|
|
|
return ore_get_rw_state(layout, comps, true, 0, 0, ios);
|
2010-11-17 02:09:58 +08:00
|
|
|
}
|
2011-08-07 10:22:06 +08:00
|
|
|
EXPORT_SYMBOL(ore_get_io_state);
|
2010-11-17 02:09:58 +08:00
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
void ore_put_io_state(struct ore_io_state *ios)
|
2008-10-28 00:27:55 +08:00
|
|
|
{
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
if (ios) {
|
|
|
|
unsigned i;
|
2008-10-28 00:27:55 +08:00
|
|
|
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
for (i = 0; i < ios->numdevs; i++) {
|
2011-08-07 10:26:31 +08:00
|
|
|
struct ore_per_dev_state *per_dev = &ios->per_dev[i];
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
|
|
|
if (per_dev->or)
|
|
|
|
osd_end_request(per_dev->or);
|
|
|
|
if (per_dev->bio)
|
|
|
|
bio_put(per_dev->bio);
|
|
|
|
}
|
|
|
|
|
|
|
|
kfree(ios);
|
2008-10-28 00:27:55 +08:00
|
|
|
}
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
}
|
2011-08-07 10:22:06 +08:00
|
|
|
EXPORT_SYMBOL(ore_put_io_state);
|
2008-10-28 00:27:55 +08:00
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
static void _sync_done(struct ore_io_state *ios, void *p)
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
{
|
|
|
|
struct completion *waiting = p;
|
2008-10-28 00:27:55 +08:00
|
|
|
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
complete(waiting);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void _last_io(struct kref *kref)
|
|
|
|
{
|
2011-08-07 10:26:31 +08:00
|
|
|
struct ore_io_state *ios = container_of(
|
|
|
|
kref, struct ore_io_state, kref);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
|
|
|
ios->done(ios, ios->private);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void _done_io(struct osd_request *or, void *p)
|
|
|
|
{
|
2011-08-07 10:26:31 +08:00
|
|
|
struct ore_io_state *ios = p;
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
|
|
|
kref_put(&ios->kref, _last_io);
|
|
|
|
}
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
static int ore_io_execute(struct ore_io_state *ios)
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
{
|
|
|
|
DECLARE_COMPLETION_ONSTACK(wait);
|
|
|
|
bool sync = (ios->done == NULL);
|
|
|
|
int i, ret;
|
|
|
|
|
|
|
|
if (sync) {
|
|
|
|
ios->done = _sync_done;
|
|
|
|
ios->private = &wait;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (i = 0; i < ios->numdevs; i++) {
|
|
|
|
struct osd_request *or = ios->per_dev[i].or;
|
|
|
|
if (unlikely(!or))
|
|
|
|
continue;
|
|
|
|
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
ret = osd_finalize_request(or, 0, _ios_cred(ios, i), NULL);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
if (unlikely(ret)) {
|
2011-08-07 10:26:31 +08:00
|
|
|
ORE_DBGMSG("Failed to osd_finalize_request() => %d\n",
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
ret);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
kref_init(&ios->kref);
|
|
|
|
|
|
|
|
for (i = 0; i < ios->numdevs; i++) {
|
|
|
|
struct osd_request *or = ios->per_dev[i].or;
|
|
|
|
if (unlikely(!or))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
kref_get(&ios->kref);
|
|
|
|
osd_execute_request_async(or, _done_io, ios);
|
|
|
|
}
|
|
|
|
|
|
|
|
kref_put(&ios->kref, _last_io);
|
|
|
|
ret = 0;
|
|
|
|
|
|
|
|
if (sync) {
|
|
|
|
wait_for_completion(&wait);
|
2011-08-07 10:26:31 +08:00
|
|
|
ret = ore_check_io(ios, NULL);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
}
|
2008-10-28 00:27:55 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2010-01-20 01:24:45 +08:00
|
|
|
static void _clear_bio(struct bio *bio)
|
|
|
|
{
|
|
|
|
struct bio_vec *bv;
|
|
|
|
unsigned i;
|
|
|
|
|
|
|
|
__bio_for_each_segment(bv, bio, i, 0) {
|
|
|
|
unsigned this_count = bv->bv_len;
|
|
|
|
|
|
|
|
if (likely(PAGE_SIZE == this_count))
|
|
|
|
clear_highpage(bv->bv_page);
|
|
|
|
else
|
|
|
|
zero_user(bv->bv_page, bv->bv_offset, this_count);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
int ore_check_io(struct ore_io_state *ios, u64 *resid)
|
2008-10-28 00:27:55 +08:00
|
|
|
{
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
enum osd_err_priority acumulated_osd_err = 0;
|
|
|
|
int acumulated_lin_err = 0;
|
|
|
|
int i;
|
2008-10-28 00:27:55 +08:00
|
|
|
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
for (i = 0; i < ios->numdevs; i++) {
|
|
|
|
struct osd_sense_info osi;
|
2010-01-20 01:24:45 +08:00
|
|
|
struct osd_request *or = ios->per_dev[i].or;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
if (unlikely(!or))
|
|
|
|
continue;
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
2010-01-20 01:24:45 +08:00
|
|
|
ret = osd_req_decode_sense(or, &osi);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
if (likely(!ret))
|
|
|
|
continue;
|
|
|
|
|
2010-01-20 01:24:45 +08:00
|
|
|
if (OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) {
|
|
|
|
/* start read offset passed endof file */
|
|
|
|
_clear_bio(ios->per_dev[i].bio);
|
2011-08-07 10:26:31 +08:00
|
|
|
ORE_DBGMSG("start read offset passed end of file "
|
2010-01-20 01:24:45 +08:00
|
|
|
"offset=0x%llx, length=0x%llx\n",
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
_LLU(ios->per_dev[i].offset),
|
|
|
|
_LLU(ios->per_dev[i].length));
|
2010-01-20 01:24:45 +08:00
|
|
|
|
|
|
|
continue; /* we recovered */
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
if (osi.osd_err_pri >= acumulated_osd_err) {
|
|
|
|
acumulated_osd_err = osi.osd_err_pri;
|
|
|
|
acumulated_lin_err = ret;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* TODO: raid specific residual calculations */
|
|
|
|
if (resid) {
|
|
|
|
if (likely(!acumulated_lin_err))
|
|
|
|
*resid = 0;
|
|
|
|
else
|
|
|
|
*resid = ios->length;
|
|
|
|
}
|
|
|
|
|
|
|
|
return acumulated_lin_err;
|
|
|
|
}
|
2011-08-07 10:22:06 +08:00
|
|
|
EXPORT_SYMBOL(ore_check_io);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
2010-02-08 01:18:58 +08:00
|
|
|
/*
|
|
|
|
* L - logical offset into the file
|
|
|
|
*
|
2010-02-11 19:01:39 +08:00
|
|
|
* U - The number of bytes in a stripe within a group
|
2010-02-08 01:18:58 +08:00
|
|
|
*
|
|
|
|
* U = stripe_unit * group_width
|
|
|
|
*
|
2010-02-11 19:01:39 +08:00
|
|
|
* T - The number of bytes striped within a group of component objects
|
|
|
|
* (before advancing to the next group)
|
2010-02-08 01:18:58 +08:00
|
|
|
*
|
2010-02-11 19:01:39 +08:00
|
|
|
* T = stripe_unit * group_width * group_depth
|
|
|
|
*
|
|
|
|
* S - The number of bytes striped across all component objects
|
|
|
|
* before the pattern repeats
|
|
|
|
*
|
|
|
|
* S = stripe_unit * group_width * group_depth * group_count
|
|
|
|
*
|
|
|
|
* M - The "major" (i.e., across all components) stripe number
|
|
|
|
*
|
|
|
|
* M = L / S
|
|
|
|
*
|
|
|
|
* G - Counts the groups from the beginning of the major stripe
|
|
|
|
*
|
|
|
|
* G = (L - (M * S)) / T [or (L % S) / T]
|
|
|
|
*
|
|
|
|
* H - The byte offset within the group
|
|
|
|
*
|
|
|
|
* H = (L - (M * S)) % T [or (L % S) % T]
|
|
|
|
*
|
|
|
|
* N - The "minor" (i.e., across the group) stripe number
|
|
|
|
*
|
|
|
|
* N = H / U
|
2010-02-08 01:18:58 +08:00
|
|
|
*
|
|
|
|
* C - The component index coresponding to L
|
|
|
|
*
|
2010-02-11 19:01:39 +08:00
|
|
|
* C = (H - (N * U)) / stripe_unit + G * group_width
|
|
|
|
* [or (L % U) / stripe_unit + G * group_width]
|
2010-02-08 01:18:58 +08:00
|
|
|
*
|
|
|
|
* O - The component offset coresponding to L
|
|
|
|
*
|
2010-02-11 19:01:39 +08:00
|
|
|
* O = L % stripe_unit + N * stripe_unit + M * group_depth * stripe_unit
|
2010-02-08 01:18:58 +08:00
|
|
|
*/
|
|
|
|
struct _striping_info {
|
|
|
|
u64 obj_offset;
|
2010-02-11 19:01:39 +08:00
|
|
|
u64 group_length;
|
2011-08-04 11:44:16 +08:00
|
|
|
u64 M; /* for truncate */
|
2010-02-08 01:18:58 +08:00
|
|
|
unsigned dev;
|
|
|
|
unsigned unit_off;
|
|
|
|
};
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
static void _calc_stripe_info(struct ore_layout *layout, u64 file_offset,
|
2010-02-08 01:18:58 +08:00
|
|
|
struct _striping_info *si)
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
{
|
2011-08-04 11:44:16 +08:00
|
|
|
u32 stripe_unit = layout->stripe_unit;
|
|
|
|
u32 group_width = layout->group_width;
|
|
|
|
u64 group_depth = layout->group_depth;
|
2010-02-11 19:01:39 +08:00
|
|
|
|
2010-02-08 01:18:58 +08:00
|
|
|
u32 U = stripe_unit * group_width;
|
2010-02-11 19:01:39 +08:00
|
|
|
u64 T = U * group_depth;
|
2011-08-04 11:44:16 +08:00
|
|
|
u64 S = T * layout->group_count;
|
2010-02-11 19:01:39 +08:00
|
|
|
u64 M = div64_u64(file_offset, S);
|
|
|
|
|
|
|
|
/*
|
|
|
|
G = (L - (M * S)) / T
|
|
|
|
H = (L - (M * S)) % T
|
|
|
|
*/
|
|
|
|
u64 LmodS = file_offset - M * S;
|
|
|
|
u32 G = div64_u64(LmodS, T);
|
|
|
|
u64 H = LmodS - G * T;
|
|
|
|
|
|
|
|
u32 N = div_u64(H, U);
|
|
|
|
|
|
|
|
/* "H - (N * U)" is just "H % U" so it's bound to u32 */
|
|
|
|
si->dev = (u32)(H - (N * U)) / stripe_unit + G * group_width;
|
2011-08-04 11:44:16 +08:00
|
|
|
si->dev *= layout->mirrors_p1;
|
2010-02-08 01:18:58 +08:00
|
|
|
|
2010-02-11 19:01:39 +08:00
|
|
|
div_u64_rem(file_offset, stripe_unit, &si->unit_off);
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
|
2010-02-11 19:01:39 +08:00
|
|
|
si->obj_offset = si->unit_off + (N * stripe_unit) +
|
|
|
|
(M * group_depth * stripe_unit);
|
|
|
|
|
|
|
|
si->group_length = T - H;
|
2011-08-04 11:44:16 +08:00
|
|
|
si->M = M;
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
}
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
static int _add_stripe_unit(struct ore_io_state *ios, unsigned *cur_pg,
|
|
|
|
unsigned pgbase, struct ore_per_dev_state *per_dev,
|
2010-01-29 00:24:06 +08:00
|
|
|
int cur_len)
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
{
|
2010-01-29 00:24:06 +08:00
|
|
|
unsigned pg = *cur_pg;
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
struct request_queue *q =
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
osd_request_queue(_ios_od(ios, per_dev->dev));
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
|
|
|
|
per_dev->length += cur_len;
|
|
|
|
|
|
|
|
if (per_dev->bio == NULL) {
|
|
|
|
unsigned pages_in_stripe = ios->layout->group_width *
|
|
|
|
(ios->layout->stripe_unit / PAGE_SIZE);
|
2010-01-29 00:24:06 +08:00
|
|
|
unsigned bio_size = (ios->nr_pages + pages_in_stripe) /
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
ios->layout->group_width;
|
|
|
|
|
|
|
|
per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size);
|
|
|
|
if (unlikely(!per_dev->bio)) {
|
2011-08-07 10:26:31 +08:00
|
|
|
ORE_DBGMSG("Failed to allocate BIO size=%u\n",
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
bio_size);
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
while (cur_len > 0) {
|
2010-01-29 00:24:06 +08:00
|
|
|
unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len);
|
|
|
|
unsigned added_len;
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
|
2010-01-29 00:24:06 +08:00
|
|
|
BUG_ON(ios->nr_pages <= pg);
|
|
|
|
cur_len -= pglen;
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
|
2010-01-29 00:24:06 +08:00
|
|
|
added_len = bio_add_pc_page(q, per_dev->bio, ios->pages[pg],
|
|
|
|
pglen, pgbase);
|
|
|
|
if (unlikely(pglen != added_len))
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
return -ENOMEM;
|
2010-01-29 00:24:06 +08:00
|
|
|
pgbase = 0;
|
|
|
|
++pg;
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
}
|
|
|
|
BUG_ON(cur_len);
|
|
|
|
|
2010-01-29 00:24:06 +08:00
|
|
|
*cur_pg = pg;
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
static int _prepare_one_group(struct ore_io_state *ios, u64 length,
|
2010-07-29 22:08:13 +08:00
|
|
|
struct _striping_info *si)
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
{
|
|
|
|
unsigned stripe_unit = ios->layout->stripe_unit;
|
2010-02-08 01:18:58 +08:00
|
|
|
unsigned mirrors_p1 = ios->layout->mirrors_p1;
|
2010-02-11 19:01:39 +08:00
|
|
|
unsigned devs_in_group = ios->layout->group_width * mirrors_p1;
|
2010-02-08 01:18:58 +08:00
|
|
|
unsigned dev = si->dev;
|
2010-02-11 19:01:39 +08:00
|
|
|
unsigned first_dev = dev - (dev % devs_in_group);
|
|
|
|
unsigned max_comp = ios->numdevs ? ios->numdevs - mirrors_p1 : 0;
|
|
|
|
unsigned cur_pg = ios->pages_consumed;
|
2010-01-29 00:24:06 +08:00
|
|
|
int ret = 0;
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
|
|
|
|
while (length) {
|
2011-08-07 10:26:31 +08:00
|
|
|
struct ore_per_dev_state *per_dev = &ios->per_dev[dev];
|
2010-02-08 01:18:58 +08:00
|
|
|
unsigned cur_len, page_off = 0;
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
|
|
|
|
if (!per_dev->length) {
|
2010-02-08 01:18:58 +08:00
|
|
|
per_dev->dev = dev;
|
|
|
|
if (dev < si->dev) {
|
|
|
|
per_dev->offset = si->obj_offset + stripe_unit -
|
|
|
|
si->unit_off;
|
|
|
|
cur_len = stripe_unit;
|
|
|
|
} else if (dev == si->dev) {
|
|
|
|
per_dev->offset = si->obj_offset;
|
|
|
|
cur_len = stripe_unit - si->unit_off;
|
|
|
|
page_off = si->unit_off & ~PAGE_MASK;
|
|
|
|
BUG_ON(page_off && (page_off != ios->pgbase));
|
|
|
|
} else { /* dev > si->dev */
|
|
|
|
per_dev->offset = si->obj_offset - si->unit_off;
|
|
|
|
cur_len = stripe_unit;
|
|
|
|
}
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
|
2010-07-29 22:08:13 +08:00
|
|
|
if (max_comp < dev)
|
|
|
|
max_comp = dev;
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
} else {
|
2010-02-08 01:18:58 +08:00
|
|
|
cur_len = stripe_unit;
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
}
|
2010-02-08 01:18:58 +08:00
|
|
|
if (cur_len >= length)
|
|
|
|
cur_len = length;
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
|
2010-01-29 00:24:06 +08:00
|
|
|
ret = _add_stripe_unit(ios, &cur_pg, page_off , per_dev,
|
|
|
|
cur_len);
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
if (unlikely(ret))
|
|
|
|
goto out;
|
|
|
|
|
2010-07-29 22:08:13 +08:00
|
|
|
dev += mirrors_p1;
|
|
|
|
dev = (dev % devs_in_group) + first_dev;
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
|
|
|
|
length -= cur_len;
|
|
|
|
}
|
|
|
|
out:
|
2010-02-11 19:01:39 +08:00
|
|
|
ios->numdevs = max_comp + mirrors_p1;
|
|
|
|
ios->pages_consumed = cur_pg;
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
static int _prepare_for_striping(struct ore_io_state *ios)
|
2010-02-08 01:18:58 +08:00
|
|
|
{
|
2010-02-11 19:01:39 +08:00
|
|
|
u64 length = ios->length;
|
2010-08-03 01:06:46 +08:00
|
|
|
u64 offset = ios->offset;
|
2010-02-08 01:18:58 +08:00
|
|
|
struct _striping_info si;
|
2010-02-11 19:01:39 +08:00
|
|
|
int ret = 0;
|
2010-02-08 01:18:58 +08:00
|
|
|
|
|
|
|
if (!ios->pages) {
|
|
|
|
if (ios->kern_buff) {
|
2011-08-07 10:26:31 +08:00
|
|
|
struct ore_per_dev_state *per_dev = &ios->per_dev[0];
|
2010-02-08 01:18:58 +08:00
|
|
|
|
2011-08-04 11:44:16 +08:00
|
|
|
_calc_stripe_info(ios->layout, ios->offset, &si);
|
2010-02-08 01:18:58 +08:00
|
|
|
per_dev->offset = si.obj_offset;
|
|
|
|
per_dev->dev = si.dev;
|
|
|
|
|
|
|
|
/* no cross device without page array */
|
|
|
|
BUG_ON((ios->layout->group_width > 1) &&
|
|
|
|
(si.unit_off + ios->length >
|
|
|
|
ios->layout->stripe_unit));
|
|
|
|
}
|
|
|
|
ios->numdevs = ios->layout->mirrors_p1;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2010-02-11 19:01:39 +08:00
|
|
|
while (length) {
|
2011-08-04 11:44:16 +08:00
|
|
|
_calc_stripe_info(ios->layout, offset, &si);
|
2010-08-03 01:06:46 +08:00
|
|
|
|
2010-02-11 19:01:39 +08:00
|
|
|
if (length < si.group_length)
|
|
|
|
si.group_length = length;
|
|
|
|
|
2010-07-29 22:08:13 +08:00
|
|
|
ret = _prepare_one_group(ios, si.group_length, &si);
|
2010-02-11 19:01:39 +08:00
|
|
|
if (unlikely(ret))
|
|
|
|
goto out;
|
|
|
|
|
2010-08-03 01:06:46 +08:00
|
|
|
offset += si.group_length;
|
2010-02-11 19:01:39 +08:00
|
|
|
length -= si.group_length;
|
|
|
|
}
|
|
|
|
|
|
|
|
out:
|
|
|
|
return ret;
|
2010-02-08 01:18:58 +08:00
|
|
|
}
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
int ore_create(struct ore_io_state *ios)
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
{
|
|
|
|
int i, ret;
|
|
|
|
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
for (i = 0; i < ios->comps->numdevs; i++) {
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
struct osd_request *or;
|
|
|
|
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
or = osd_start_request(_ios_od(ios, i), GFP_KERNEL);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
if (unlikely(!or)) {
|
2011-08-07 10:26:31 +08:00
|
|
|
ORE_ERR("%s: osd_start_request failed\n", __func__);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
ret = -ENOMEM;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
ios->per_dev[i].or = or;
|
|
|
|
ios->numdevs++;
|
|
|
|
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
osd_req_create_object(or, _ios_obj(ios, i));
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
}
|
2011-08-07 10:26:31 +08:00
|
|
|
ret = ore_io_execute(ios);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
|
|
|
out:
|
|
|
|
return ret;
|
|
|
|
}
|
2011-08-07 10:22:06 +08:00
|
|
|
EXPORT_SYMBOL(ore_create);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
int ore_remove(struct ore_io_state *ios)
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
{
|
|
|
|
int i, ret;
|
|
|
|
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
for (i = 0; i < ios->comps->numdevs; i++) {
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
struct osd_request *or;
|
|
|
|
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
or = osd_start_request(_ios_od(ios, i), GFP_KERNEL);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
if (unlikely(!or)) {
|
2011-08-07 10:26:31 +08:00
|
|
|
ORE_ERR("%s: osd_start_request failed\n", __func__);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
ret = -ENOMEM;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
ios->per_dev[i].or = or;
|
|
|
|
ios->numdevs++;
|
|
|
|
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
osd_req_remove_object(or, _ios_obj(ios, i));
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
}
|
2011-08-07 10:26:31 +08:00
|
|
|
ret = ore_io_execute(ios);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
|
|
|
out:
|
|
|
|
return ret;
|
|
|
|
}
|
2011-08-07 10:22:06 +08:00
|
|
|
EXPORT_SYMBOL(ore_remove);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
static int _write_mirror(struct ore_io_state *ios, int cur_comp)
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
{
|
2011-08-07 10:26:31 +08:00
|
|
|
struct ore_per_dev_state *master_dev = &ios->per_dev[cur_comp];
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
unsigned dev = ios->per_dev[cur_comp].dev;
|
|
|
|
unsigned last_comp = cur_comp + ios->layout->mirrors_p1;
|
|
|
|
int ret = 0;
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
2010-02-11 19:01:39 +08:00
|
|
|
if (ios->pages && !master_dev->length)
|
|
|
|
return 0; /* Just an empty slot */
|
|
|
|
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
for (; cur_comp < last_comp; ++cur_comp, ++dev) {
|
2011-08-07 10:26:31 +08:00
|
|
|
struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
struct osd_request *or;
|
|
|
|
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
or = osd_start_request(_ios_od(ios, dev), GFP_KERNEL);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
if (unlikely(!or)) {
|
2011-08-07 10:26:31 +08:00
|
|
|
ORE_ERR("%s: osd_start_request failed\n", __func__);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
ret = -ENOMEM;
|
|
|
|
goto out;
|
|
|
|
}
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
per_dev->or = or;
|
|
|
|
per_dev->offset = master_dev->offset;
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
2010-01-29 00:24:06 +08:00
|
|
|
if (ios->pages) {
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
struct bio *bio;
|
|
|
|
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
if (per_dev != master_dev) {
|
exofs: Multi-device mirror support
This patch changes on-disk format, it is accompanied with a parallel
patch to mkfs.exofs that enables multi-device capabilities.
After this patch, old exofs will refuse to mount a new formatted FS and
new exofs will refuse an old format. This is done by moving the magic
field offset inside the FSCB. A new FSCB *version* field was added. In
the future, exofs will refuse to mount unmatched FSCB version. To
up-grade or down-grade an exofs one must use mkfs.exofs --upgrade option
before mounting.
Introduced, a new object that contains a *device-table*. This object
contains the default *data-map* and a linear array of devices
information, which identifies the devices used in the filesystem. This
object is only written to offline by mkfs.exofs. This is why it is kept
separate from the FSCB, since the later is written to while mounted.
Same partition number, same object number is used on all devices only
the device varies.
* define the new format, then load the device table on mount time make
sure every thing is supported.
* Change I/O engine to now support Mirror IO, .i.e write same data
to multiple devices, read from a random device to spread the
read-load from multiple clients (TODO: stripe read)
Implementation notes:
A few points introduced in previous patch should be mentioned here:
* Special care was made so absolutlly all operation that have any chance
of failing are done before any osd-request is executed. This is to
minimize the need for a data consistency recovery, to only real IO
errors.
* Each IO state has a kref. It starts at 1, any osd-request executed
will increment the kref, finally when all are executed the first ref
is dropped. At IO-done, each request completion decrements the kref,
the last one to return executes the internal _last_io() routine.
_last_io() will call the registered io_state_done. On sync mode a
caller does not supply a done method, indicating a synchronous
request, the caller is put to sleep and a special io_state_done is
registered that will awaken the caller. Though also in sync mode all
operations are executed in parallel.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-16 22:03:05 +08:00
|
|
|
bio = bio_kmalloc(GFP_KERNEL,
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
master_dev->bio->bi_max_vecs);
|
exofs: Multi-device mirror support
This patch changes on-disk format, it is accompanied with a parallel
patch to mkfs.exofs that enables multi-device capabilities.
After this patch, old exofs will refuse to mount a new formatted FS and
new exofs will refuse an old format. This is done by moving the magic
field offset inside the FSCB. A new FSCB *version* field was added. In
the future, exofs will refuse to mount unmatched FSCB version. To
up-grade or down-grade an exofs one must use mkfs.exofs --upgrade option
before mounting.
Introduced, a new object that contains a *device-table*. This object
contains the default *data-map* and a linear array of devices
information, which identifies the devices used in the filesystem. This
object is only written to offline by mkfs.exofs. This is why it is kept
separate from the FSCB, since the later is written to while mounted.
Same partition number, same object number is used on all devices only
the device varies.
* define the new format, then load the device table on mount time make
sure every thing is supported.
* Change I/O engine to now support Mirror IO, .i.e write same data
to multiple devices, read from a random device to spread the
read-load from multiple clients (TODO: stripe read)
Implementation notes:
A few points introduced in previous patch should be mentioned here:
* Special care was made so absolutlly all operation that have any chance
of failing are done before any osd-request is executed. This is to
minimize the need for a data consistency recovery, to only real IO
errors.
* Each IO state has a kref. It starts at 1, any osd-request executed
will increment the kref, finally when all are executed the first ref
is dropped. At IO-done, each request completion decrements the kref,
the last one to return executes the internal _last_io() routine.
_last_io() will call the registered io_state_done. On sync mode a
caller does not supply a done method, indicating a synchronous
request, the caller is put to sleep and a special io_state_done is
registered that will awaken the caller. Though also in sync mode all
operations are executed in parallel.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-16 22:03:05 +08:00
|
|
|
if (unlikely(!bio)) {
|
2011-08-07 10:26:31 +08:00
|
|
|
ORE_DBGMSG(
|
2010-08-07 18:30:03 +08:00
|
|
|
"Failed to allocate BIO size=%u\n",
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
master_dev->bio->bi_max_vecs);
|
exofs: Multi-device mirror support
This patch changes on-disk format, it is accompanied with a parallel
patch to mkfs.exofs that enables multi-device capabilities.
After this patch, old exofs will refuse to mount a new formatted FS and
new exofs will refuse an old format. This is done by moving the magic
field offset inside the FSCB. A new FSCB *version* field was added. In
the future, exofs will refuse to mount unmatched FSCB version. To
up-grade or down-grade an exofs one must use mkfs.exofs --upgrade option
before mounting.
Introduced, a new object that contains a *device-table*. This object
contains the default *data-map* and a linear array of devices
information, which identifies the devices used in the filesystem. This
object is only written to offline by mkfs.exofs. This is why it is kept
separate from the FSCB, since the later is written to while mounted.
Same partition number, same object number is used on all devices only
the device varies.
* define the new format, then load the device table on mount time make
sure every thing is supported.
* Change I/O engine to now support Mirror IO, .i.e write same data
to multiple devices, read from a random device to spread the
read-load from multiple clients (TODO: stripe read)
Implementation notes:
A few points introduced in previous patch should be mentioned here:
* Special care was made so absolutlly all operation that have any chance
of failing are done before any osd-request is executed. This is to
minimize the need for a data consistency recovery, to only real IO
errors.
* Each IO state has a kref. It starts at 1, any osd-request executed
will increment the kref, finally when all are executed the first ref
is dropped. At IO-done, each request completion decrements the kref,
the last one to return executes the internal _last_io() routine.
_last_io() will call the registered io_state_done. On sync mode a
caller does not supply a done method, indicating a synchronous
request, the caller is put to sleep and a special io_state_done is
registered that will awaken the caller. Though also in sync mode all
operations are executed in parallel.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-16 22:03:05 +08:00
|
|
|
ret = -ENOMEM;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
__bio_clone(bio, master_dev->bio);
|
exofs: Multi-device mirror support
This patch changes on-disk format, it is accompanied with a parallel
patch to mkfs.exofs that enables multi-device capabilities.
After this patch, old exofs will refuse to mount a new formatted FS and
new exofs will refuse an old format. This is done by moving the magic
field offset inside the FSCB. A new FSCB *version* field was added. In
the future, exofs will refuse to mount unmatched FSCB version. To
up-grade or down-grade an exofs one must use mkfs.exofs --upgrade option
before mounting.
Introduced, a new object that contains a *device-table*. This object
contains the default *data-map* and a linear array of devices
information, which identifies the devices used in the filesystem. This
object is only written to offline by mkfs.exofs. This is why it is kept
separate from the FSCB, since the later is written to while mounted.
Same partition number, same object number is used on all devices only
the device varies.
* define the new format, then load the device table on mount time make
sure every thing is supported.
* Change I/O engine to now support Mirror IO, .i.e write same data
to multiple devices, read from a random device to spread the
read-load from multiple clients (TODO: stripe read)
Implementation notes:
A few points introduced in previous patch should be mentioned here:
* Special care was made so absolutlly all operation that have any chance
of failing are done before any osd-request is executed. This is to
minimize the need for a data consistency recovery, to only real IO
errors.
* Each IO state has a kref. It starts at 1, any osd-request executed
will increment the kref, finally when all are executed the first ref
is dropped. At IO-done, each request completion decrements the kref,
the last one to return executes the internal _last_io() routine.
_last_io() will call the registered io_state_done. On sync mode a
caller does not supply a done method, indicating a synchronous
request, the caller is put to sleep and a special io_state_done is
registered that will awaken the caller. Though also in sync mode all
operations are executed in parallel.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-16 22:03:05 +08:00
|
|
|
bio->bi_bdev = NULL;
|
|
|
|
bio->bi_next = NULL;
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
per_dev->length = master_dev->length;
|
|
|
|
per_dev->bio = bio;
|
|
|
|
per_dev->dev = dev;
|
exofs: Multi-device mirror support
This patch changes on-disk format, it is accompanied with a parallel
patch to mkfs.exofs that enables multi-device capabilities.
After this patch, old exofs will refuse to mount a new formatted FS and
new exofs will refuse an old format. This is done by moving the magic
field offset inside the FSCB. A new FSCB *version* field was added. In
the future, exofs will refuse to mount unmatched FSCB version. To
up-grade or down-grade an exofs one must use mkfs.exofs --upgrade option
before mounting.
Introduced, a new object that contains a *device-table*. This object
contains the default *data-map* and a linear array of devices
information, which identifies the devices used in the filesystem. This
object is only written to offline by mkfs.exofs. This is why it is kept
separate from the FSCB, since the later is written to while mounted.
Same partition number, same object number is used on all devices only
the device varies.
* define the new format, then load the device table on mount time make
sure every thing is supported.
* Change I/O engine to now support Mirror IO, .i.e write same data
to multiple devices, read from a random device to spread the
read-load from multiple clients (TODO: stripe read)
Implementation notes:
A few points introduced in previous patch should be mentioned here:
* Special care was made so absolutlly all operation that have any chance
of failing are done before any osd-request is executed. This is to
minimize the need for a data consistency recovery, to only real IO
errors.
* Each IO state has a kref. It starts at 1, any osd-request executed
will increment the kref, finally when all are executed the first ref
is dropped. At IO-done, each request completion decrements the kref,
the last one to return executes the internal _last_io() routine.
_last_io() will call the registered io_state_done. On sync mode a
caller does not supply a done method, indicating a synchronous
request, the caller is put to sleep and a special io_state_done is
registered that will awaken the caller. Though also in sync mode all
operations are executed in parallel.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-16 22:03:05 +08:00
|
|
|
} else {
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
bio = master_dev->bio;
|
|
|
|
/* FIXME: bio_set_dir() */
|
2010-08-08 00:20:39 +08:00
|
|
|
bio->bi_rw |= REQ_WRITE;
|
exofs: Multi-device mirror support
This patch changes on-disk format, it is accompanied with a parallel
patch to mkfs.exofs that enables multi-device capabilities.
After this patch, old exofs will refuse to mount a new formatted FS and
new exofs will refuse an old format. This is done by moving the magic
field offset inside the FSCB. A new FSCB *version* field was added. In
the future, exofs will refuse to mount unmatched FSCB version. To
up-grade or down-grade an exofs one must use mkfs.exofs --upgrade option
before mounting.
Introduced, a new object that contains a *device-table*. This object
contains the default *data-map* and a linear array of devices
information, which identifies the devices used in the filesystem. This
object is only written to offline by mkfs.exofs. This is why it is kept
separate from the FSCB, since the later is written to while mounted.
Same partition number, same object number is used on all devices only
the device varies.
* define the new format, then load the device table on mount time make
sure every thing is supported.
* Change I/O engine to now support Mirror IO, .i.e write same data
to multiple devices, read from a random device to spread the
read-load from multiple clients (TODO: stripe read)
Implementation notes:
A few points introduced in previous patch should be mentioned here:
* Special care was made so absolutlly all operation that have any chance
of failing are done before any osd-request is executed. This is to
minimize the need for a data consistency recovery, to only real IO
errors.
* Each IO state has a kref. It starts at 1, any osd-request executed
will increment the kref, finally when all are executed the first ref
is dropped. At IO-done, each request completion decrements the kref,
the last one to return executes the internal _last_io() routine.
_last_io() will call the registered io_state_done. On sync mode a
caller does not supply a done method, indicating a synchronous
request, the caller is put to sleep and a special io_state_done is
registered that will awaken the caller. Though also in sync mode all
operations are executed in parallel.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-16 22:03:05 +08:00
|
|
|
}
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
osd_req_write(or, _ios_obj(ios, dev), per_dev->offset,
|
|
|
|
bio, per_dev->length);
|
2011-08-07 10:26:31 +08:00
|
|
|
ORE_DBGMSG("write(0x%llx) offset=0x%llx "
|
2009-12-16 01:34:17 +08:00
|
|
|
"length=0x%llx dev=%d\n",
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
_LLU(_ios_obj(ios, dev)->id),
|
|
|
|
_LLU(per_dev->offset),
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
_LLU(per_dev->length), dev);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
} else if (ios->kern_buff) {
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
ret = osd_req_write_kern(or, _ios_obj(ios, dev),
|
|
|
|
per_dev->offset,
|
|
|
|
ios->kern_buff, ios->length);
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
if (unlikely(ret))
|
|
|
|
goto out;
|
2011-08-07 10:26:31 +08:00
|
|
|
ORE_DBGMSG2("write_kern(0x%llx) offset=0x%llx "
|
2009-12-16 01:34:17 +08:00
|
|
|
"length=0x%llx dev=%d\n",
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
_LLU(_ios_obj(ios, dev)->id),
|
|
|
|
_LLU(per_dev->offset),
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
_LLU(ios->length), dev);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
} else {
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
osd_req_set_attributes(or, _ios_obj(ios, dev));
|
2011-08-07 10:26:31 +08:00
|
|
|
ORE_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n",
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
_LLU(_ios_obj(ios, dev)->id),
|
|
|
|
ios->out_attr_len, dev);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
if (ios->out_attr)
|
|
|
|
osd_req_add_set_attr_list(or, ios->out_attr,
|
|
|
|
ios->out_attr_len);
|
|
|
|
|
|
|
|
if (ios->in_attr)
|
|
|
|
osd_req_add_get_attr_list(or, ios->in_attr,
|
|
|
|
ios->in_attr_len);
|
2008-10-28 00:27:55 +08:00
|
|
|
}
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
|
|
|
out:
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
int ore_write(struct ore_io_state *ios)
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
{
|
|
|
|
int i;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = _prepare_for_striping(ios);
|
|
|
|
if (unlikely(ret))
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
|
2011-08-07 10:26:31 +08:00
|
|
|
ret = _write_mirror(ios, i);
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
if (unlikely(ret))
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
ret = ore_io_execute(ios);
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
return ret;
|
|
|
|
}
|
2011-08-07 10:22:06 +08:00
|
|
|
EXPORT_SYMBOL(ore_write);
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
static int _read_mirror(struct ore_io_state *ios, unsigned cur_comp)
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
{
|
2010-02-01 17:37:30 +08:00
|
|
|
struct osd_request *or;
|
2011-08-07 10:26:31 +08:00
|
|
|
struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
struct osd_obj_id *obj = _ios_obj(ios, cur_comp);
|
|
|
|
unsigned first_dev = (unsigned)obj->id;
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
2010-02-11 19:01:39 +08:00
|
|
|
if (ios->pages && !per_dev->length)
|
|
|
|
return 0; /* Just an empty slot */
|
|
|
|
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1;
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
or = osd_start_request(_ios_od(ios, first_dev), GFP_KERNEL);
|
2010-02-01 17:37:30 +08:00
|
|
|
if (unlikely(!or)) {
|
2011-08-07 10:26:31 +08:00
|
|
|
ORE_ERR("%s: osd_start_request failed\n", __func__);
|
2010-02-01 17:37:30 +08:00
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
per_dev->or = or;
|
|
|
|
|
2010-01-29 00:24:06 +08:00
|
|
|
if (ios->pages) {
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
osd_req_read(or, obj, per_dev->offset,
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
per_dev->bio, per_dev->length);
|
2011-08-07 10:26:31 +08:00
|
|
|
ORE_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx"
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
" dev=%d\n", _LLU(obj->id),
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
_LLU(per_dev->offset), _LLU(per_dev->length),
|
2010-02-01 17:37:30 +08:00
|
|
|
first_dev);
|
|
|
|
} else if (ios->kern_buff) {
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
int ret = osd_req_read_kern(or, obj, per_dev->offset,
|
2010-02-01 17:37:30 +08:00
|
|
|
ios->kern_buff, ios->length);
|
2011-08-07 10:26:31 +08:00
|
|
|
ORE_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
|
2010-02-01 17:37:30 +08:00
|
|
|
"length=0x%llx dev=%d ret=>%d\n",
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
_LLU(obj->id), _LLU(per_dev->offset),
|
2010-02-01 17:37:30 +08:00
|
|
|
_LLU(ios->length), first_dev, ret);
|
|
|
|
if (unlikely(ret))
|
|
|
|
return ret;
|
|
|
|
} else {
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
osd_req_get_attributes(or, obj);
|
2011-08-07 10:26:31 +08:00
|
|
|
ORE_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n",
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
_LLU(obj->id),
|
|
|
|
ios->in_attr_len, first_dev);
|
2010-02-01 17:37:30 +08:00
|
|
|
}
|
|
|
|
if (ios->out_attr)
|
|
|
|
osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len);
|
2008-10-28 00:27:55 +08:00
|
|
|
|
2010-02-01 17:37:30 +08:00
|
|
|
if (ios->in_attr)
|
|
|
|
osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len);
|
2008-10-28 00:27:55 +08:00
|
|
|
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
int ore_read(struct ore_io_state *ios)
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
{
|
|
|
|
int i;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = _prepare_for_striping(ios);
|
|
|
|
if (unlikely(ret))
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
|
2011-08-07 10:26:31 +08:00
|
|
|
ret = _read_mirror(ios, i);
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
if (unlikely(ret))
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
ret = ore_io_execute(ios);
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
return ret;
|
2008-10-28 00:27:55 +08:00
|
|
|
}
|
2011-08-07 10:22:06 +08:00
|
|
|
EXPORT_SYMBOL(ore_read);
|
2008-10-28 00:27:55 +08:00
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
int extract_attr_from_ios(struct ore_io_state *ios, struct osd_attr *attr)
|
2008-10-28 00:27:55 +08:00
|
|
|
{
|
|
|
|
struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
|
|
|
|
void *iter = NULL;
|
|
|
|
int nelem;
|
|
|
|
|
|
|
|
do {
|
|
|
|
nelem = 1;
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
osd_req_decode_get_attr_list(ios->per_dev[0].or,
|
|
|
|
&cur_attr, &nelem, &iter);
|
2008-10-28 00:27:55 +08:00
|
|
|
if ((cur_attr.attr_page == attr->attr_page) &&
|
|
|
|
(cur_attr.attr_id == attr->attr_id)) {
|
|
|
|
attr->len = cur_attr.len;
|
|
|
|
attr->val_ptr = cur_attr.val_ptr;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
} while (iter);
|
|
|
|
|
|
|
|
return -EIO;
|
|
|
|
}
|
2011-08-07 10:22:06 +08:00
|
|
|
EXPORT_SYMBOL(extract_attr_from_ios);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
static int _truncate_mirrors(struct ore_io_state *ios, unsigned cur_comp,
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
struct osd_attr *attr)
|
|
|
|
{
|
|
|
|
int last_comp = cur_comp + ios->layout->mirrors_p1;
|
|
|
|
|
|
|
|
for (; cur_comp < last_comp; ++cur_comp) {
|
2011-08-07 10:26:31 +08:00
|
|
|
struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
struct osd_request *or;
|
|
|
|
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
or = osd_start_request(_ios_od(ios, cur_comp), GFP_KERNEL);
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
if (unlikely(!or)) {
|
2011-08-07 10:26:31 +08:00
|
|
|
ORE_ERR("%s: osd_start_request failed\n", __func__);
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
per_dev->or = or;
|
|
|
|
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
osd_req_set_attributes(or, _ios_obj(ios, cur_comp));
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
osd_req_add_set_attr_list(or, attr, 1);
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-08-04 11:44:16 +08:00
|
|
|
struct _trunc_info {
|
|
|
|
struct _striping_info si;
|
|
|
|
u64 prev_group_obj_off;
|
|
|
|
u64 next_group_obj_off;
|
|
|
|
|
|
|
|
unsigned first_group_dev;
|
|
|
|
unsigned nex_group_dev;
|
|
|
|
unsigned max_devs;
|
|
|
|
};
|
|
|
|
|
2011-09-24 04:42:43 +08:00
|
|
|
static void _calc_trunk_info(struct ore_layout *layout, u64 file_offset,
|
|
|
|
struct _trunc_info *ti)
|
2011-08-04 11:44:16 +08:00
|
|
|
{
|
|
|
|
unsigned stripe_unit = layout->stripe_unit;
|
|
|
|
|
|
|
|
_calc_stripe_info(layout, file_offset, &ti->si);
|
|
|
|
|
|
|
|
ti->prev_group_obj_off = ti->si.M * stripe_unit;
|
|
|
|
ti->next_group_obj_off = ti->si.M ? (ti->si.M - 1) * stripe_unit : 0;
|
|
|
|
|
|
|
|
ti->first_group_dev = ti->si.dev - (ti->si.dev % layout->group_width);
|
|
|
|
ti->nex_group_dev = ti->first_group_dev + layout->group_width;
|
|
|
|
ti->max_devs = layout->group_width * layout->group_count;
|
|
|
|
}
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
int ore_truncate(struct ore_layout *layout, struct ore_components *comps,
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
u64 size)
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
{
|
2011-08-07 10:26:31 +08:00
|
|
|
struct ore_io_state *ios;
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
struct exofs_trunc_attr {
|
|
|
|
struct osd_attr attr;
|
|
|
|
__be64 newsize;
|
|
|
|
} *size_attrs;
|
2011-08-04 11:44:16 +08:00
|
|
|
struct _trunc_info ti;
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
int i, ret;
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
ret = ore_get_io_state(layout, comps, &ios);
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
if (unlikely(ret))
|
|
|
|
return ret;
|
|
|
|
|
2011-08-04 11:44:16 +08:00
|
|
|
_calc_trunk_info(ios->layout, size, &ti);
|
|
|
|
|
|
|
|
size_attrs = kcalloc(ti.max_devs, sizeof(*size_attrs),
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
GFP_KERNEL);
|
|
|
|
if (unlikely(!size_attrs)) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto out;
|
|
|
|
}
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
ios->numdevs = ios->comps->numdevs;
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
2011-08-04 11:44:16 +08:00
|
|
|
for (i = 0; i < ti.max_devs; ++i) {
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
struct exofs_trunc_attr *size_attr = &size_attrs[i];
|
|
|
|
u64 obj_size;
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
2011-08-04 11:44:16 +08:00
|
|
|
if (i < ti.first_group_dev)
|
|
|
|
obj_size = ti.prev_group_obj_off;
|
|
|
|
else if (i >= ti.nex_group_dev)
|
|
|
|
obj_size = ti.next_group_obj_off;
|
|
|
|
else if (i < ti.si.dev) /* dev within this group */
|
|
|
|
obj_size = ti.si.obj_offset +
|
|
|
|
ios->layout->stripe_unit - ti.si.unit_off;
|
|
|
|
else if (i == ti.si.dev)
|
|
|
|
obj_size = ti.si.obj_offset;
|
|
|
|
else /* i > ti.dev */
|
|
|
|
obj_size = ti.si.obj_offset - ti.si.unit_off;
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
size_attr->newsize = cpu_to_be64(obj_size);
|
|
|
|
size_attr->attr = g_attr_logical_length;
|
|
|
|
size_attr->attr.val_ptr = &size_attr->newsize;
|
|
|
|
|
2011-08-07 10:26:31 +08:00
|
|
|
ORE_DBGMSG("trunc(0x%llx) obj_offset=0x%llx dev=%d\n",
|
exofs: ios: Move to a per inode components & device-table
Exofs raid engine was saving on memory space by having a single layout-info,
single pid, and a single device-table, global to the filesystem. Then passing
a credential and object_id info at the io_state level, private for each
inode. It would also devise this contraption of rotating the device table
view for each inode->ino to spread out the device usage.
This is not compatible with the pnfs-objects standard, demanding that
each inode can have it's own layout-info, device-table, and each object
component it's own pid, oid and creds.
So: Bring exofs raid engine to be usable for generic pnfs-objects use by:
* Define an exofs_comp structure that holds obj_id and credential info.
* Break up exofs_layout struct to an exofs_components structure that holds a
possible array of exofs_comp and the array of devices + the size of the
arrays.
* Add a "comps" parameter to get_io_state() that specifies the ids creds
and device array to use for each IO.
This enables to keep the layout global, but the device-table view, creds
and IDs at the inode level. It only adds two 64bit to each inode, since
some of these members already existed in another form.
* ios raid engine now access layout-info and comps-info through the passed
pointers. Everything is pre-prepared by caller for generic access of
these structures and arrays.
At the exofs Level:
* Super block holds an exofs_components struct that holds the device
array, previously in layout. The devices there are in device-table
order. The device-array is twice bigger and repeats the device-table
twice so now each inode's device array can point to a random device
and have a round-robin view of the table, making it compatible to
previous exofs versions.
* Each inode has an exofs_components struct that is initialized at
load time, with it's own view of the device table IDs and creds.
When doing IO this gets passed to the io_state together with the
layout.
While preforming this change. Bugs where found where credentials with the
wrong IDs where used to access the different SB objects (super.c). As well
as some dead code. It was never noticed because the target we use does not
check the credentials.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2011-08-06 06:06:04 +08:00
|
|
|
_LLU(comps->comps->obj.id), _LLU(obj_size), i);
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1,
|
|
|
|
&size_attr->attr);
|
|
|
|
if (unlikely(ret))
|
|
|
|
goto out;
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
}
|
2011-08-07 10:26:31 +08:00
|
|
|
ret = ore_io_execute(ios);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
|
|
|
|
out:
|
exofs: RAID0 support
We now support striping over mirror devices. Including variable sized
stripe_unit.
Some limits:
* stripe_unit must be a multiple of PAGE_SIZE
* stripe_unit * stripe_count is maximum upto 32-bit (4Gb)
Tested RAID0 over mirrors, RAID0 only, mirrors only. All check.
Design notes:
* I'm not using a vectored raid-engine mechanism yet. Following the
pnfs-objects-layout data-map structure, "Mirror" is just a private
case of "group_width" == 1, and RAID0 is a private case of
"Mirrors" == 1. The performance lose of the general case over the
particular special case optimization is totally negligible, also
considering the extra code size.
* In general I added a prepare_stripes() stage that divides the
to-be-io pages to the participating devices, the previous
exofs_ios_write/read, now becomes _write/read_mirrors and a new
write/read upper layer loops on all devices calling
_write/read_mirrors. Effectively the prepare_stripes stage is the all
secret.
Also truncate need fixing to accommodate for striping.
* In a RAID0 arrangement, in a regular usage scenario, if all inode
layouts will start at the same device, the small files fill up the
first device and the later devices stay empty, the farther the device
the emptier it is.
To fix that, each inode will start at a different stripe_unit,
according to it's obj_id modulus number-of-stripe-units. And
will then span all stripe-units in the same incrementing order
wrapping back to the beginning of the device table. We call it
a stripe-units moving window.
Special consideration was taken to keep all devices in a mirror
arrangement identical. So a broken osd-device could just be cloned
from one of the mirrors and no FS scrubbing is needed. (We do that
by rotating stripe-unit at a time and not a single device at a time.)
TODO:
We no longer verify object_length == inode->i_size in exofs_iget.
(since i_size is stripped on multiple objects now).
I should introduce a multiple-device attribute reading, and use
it in exofs_iget.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2010-02-01 19:35:51 +08:00
|
|
|
kfree(size_attrs);
|
2011-08-07 10:26:31 +08:00
|
|
|
ore_put_io_state(ios);
|
exofs: Move all operations to an io_engine
In anticipation for multi-device operations, we separate osd operations
into an abstract I/O API. Currently only one device is used but later
when adding more devices, we will drive all devices in parallel according
to a "data_map" that describes how data is arranged on multiple devices.
The file system level operates, like before, as if there is one object
(inode-number) and an i_size. The io engine will split this to the same
object-number but on multiple device.
At first we introduce Mirror (raid 1) layout. But at the final outcome
we intend to fully implement the pNFS-Objects data-map, including
raid 0,4,5,6 over mirrored devices, over multiple device-groups. And
more. See: http://tools.ietf.org/html/draft-ietf-nfsv4-pnfs-obj-12
* Define an io_state based API for accessing osd storage devices
in an abstract way.
Usage:
First a caller allocates an io state with:
exofs_get_io_state(struct exofs_sb_info *sbi,
struct exofs_io_state** ios);
Then calles one of:
exofs_sbi_create(struct exofs_io_state *ios);
exofs_sbi_remove(struct exofs_io_state *ios);
exofs_sbi_write(struct exofs_io_state *ios);
exofs_sbi_read(struct exofs_io_state *ios);
exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
And when done
exofs_put_io_state(struct exofs_io_state *ios);
* Convert all source files to use this new API
* Convert from bio_alloc to bio_kmalloc
* In io engine we make use of the now fixed osd_req_decode_sense
There are no functional changes or on disk additions after this patch.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
2009-11-08 20:54:08 +08:00
|
|
|
return ret;
|
|
|
|
}
|
2011-08-07 10:22:06 +08:00
|
|
|
EXPORT_SYMBOL(ore_truncate);
|
2011-05-16 20:26:47 +08:00
|
|
|
|
|
|
|
const struct osd_attr g_attr_logical_length = ATTR_DEF(
|
|
|
|
OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
|
2011-08-07 10:22:06 +08:00
|
|
|
EXPORT_SYMBOL(g_attr_logical_length);
|